Re: wfcqueue (in Userspace RCU) for Linux kernel (for epoll)

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

* Eric Wong (normalperson@yhbt.net) wrote:
> Hi, I'm looking to reduce contention for the ep->lock spin lock in epoll.
> 
> I came across wfcqueue in Userspace RCU and am wondering if there's any
> reason (other that lack of developer time/users) it hasn't been adapted
> for the Linux kernel.
> 
> I'd be happy to do the work if it's suitable (and omit parts which
> may not be, such as busy wait).

Hi Eric,

I'm glad you're interested in our data structure!

I actually have a port ready, I just lacked the time for thorough
testing and also lacked users of the API at kernel-level. It looks like
now would be a good time to send the patch your way as a RFC!

> Using a mutex lock (while in process context) for dequeue works for
> epoll.  I mainly want a lock-free enqueue for ep_poll_callback since
> that may be called in IRQ context.

You'd get even better than this: wait-free enqueue, with a simple xchg
and store.

> 
> I suppose I can also use the llist LIFO in ep_poll_callback and
> reverse that while in process context, too...

Yeah, but that seems more expensive than just having the list in the
right order already. ;)

Patch incoming in a separate post,

Thanks,

Mathieu

> 
> 
> ref: http://git.lttng.org/userspace-rcu.git

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com